| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2014 x265 project |
| 3 | * |
| 4 | * Authors: Steve Borho <steve@borho.org> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
| 19 | * |
| 20 | * This program is also available under a commercial proprietary license. |
| 21 | * For more information, contact us at license @ x265.com. |
| 22 | *****************************************************************************/ |
| 23 | |
| 24 | #ifndef X265_CUDATA_H |
| 25 | #define X265_CUDATA_H |
| 26 | |
| 27 | #include "common.h" |
| 28 | #include "slice.h" |
| 29 | #include "mv.h" |
| 30 | |
| 31 | namespace x265 { |
| 32 | // private namespace |
| 33 | |
| 34 | class FrameData; |
| 35 | class Slice; |
| 36 | struct TUEntropyCodingParameters; |
| 37 | struct CUDataMemPool; |
| 38 | |
| 39 | enum PartSize |
| 40 | { |
| 41 | SIZE_2Nx2N, // symmetric motion partition, 2Nx2N |
| 42 | SIZE_2NxN, // symmetric motion partition, 2Nx N |
| 43 | SIZE_Nx2N, // symmetric motion partition, Nx2N |
| 44 | SIZE_NxN, // symmetric motion partition, Nx N |
| 45 | SIZE_2NxnU, // asymmetric motion partition, 2Nx( N/2) + 2Nx(3N/2) |
| 46 | SIZE_2NxnD, // asymmetric motion partition, 2Nx(3N/2) + 2Nx( N/2) |
| 47 | SIZE_nLx2N, // asymmetric motion partition, ( N/2)x2N + (3N/2)x2N |
| 48 | SIZE_nRx2N, // asymmetric motion partition, (3N/2)x2N + ( N/2)x2N |
| 49 | SIZE_NONE = 15 |
| 50 | }; |
| 51 | |
| 52 | enum PredMode |
| 53 | { |
| 54 | MODE_INTER, |
| 55 | MODE_INTRA, |
| 56 | MODE_NONE = 15 |
| 57 | }; |
| 58 | |
| 59 | // motion vector predictor direction used in AMVP |
| 60 | enum MVP_DIR |
| 61 | { |
| 62 | MD_LEFT = 0, // MVP of left block |
| 63 | MD_ABOVE, // MVP of above block |
| 64 | MD_ABOVE_RIGHT, // MVP of above right block |
| 65 | MD_BELOW_LEFT, // MVP of below left block |
| 66 | MD_ABOVE_LEFT // MVP of above left block |
| 67 | }; |
| 68 | |
| 69 | struct CUGeom |
| 70 | { |
| 71 | enum { |
| 72 | INTRA = 1<<0, // CU is intra predicted |
| 73 | PRESENT = 1<<1, // CU is not completely outside the frame |
| 74 | SPLIT_MANDATORY = 1<<2, // CU split is mandatory if CU is inside frame and can be split |
| 75 | LEAF = 1<<3, // CU is a leaf node of the CTU |
| 76 | SPLIT = 1<<4, // CU is currently split in four child CUs. |
| 77 | }; |
| 78 | |
| 79 | // (1 + 4 + 16 + 64) = 85. |
| 80 | enum { MAX_GEOMS = 85 }; |
| 81 | |
| 82 | uint32_t log2CUSize; // Log of the CU size. |
| 83 | uint32_t childOffset; // offset of the first child CU from current CU |
| 84 | uint32_t encodeIdx; // Encoding index of this CU in terms of 4x4 blocks. |
| 85 | uint32_t numPartitions; // Number of 4x4 blocks in the CU |
| 86 | uint32_t depth; // depth of this CU relative from CTU |
| 87 | uint32_t flags; // CU flags. |
| 88 | }; |
| 89 | |
| 90 | struct MVField |
| 91 | { |
| 92 | MV mv; |
| 93 | int refIdx; |
| 94 | }; |
| 95 | |
| 96 | typedef void(*cucopy_t)(uint8_t* dst, uint8_t* src); // dst and src are aligned to MIN(size, 32) |
| 97 | typedef void(*cubcast_t)(uint8_t* dst, uint8_t val); // dst is aligned to MIN(size, 32) |
| 98 | |
| 99 | // Partition count table, index represents partitioning mode. |
| 100 | const uint32_t nbPartsTable[8] = { 1, 2, 2, 4, 2, 2, 2, 2 }; |
| 101 | |
| 102 | // Holds part data for a CU of a given size, from an 8x8 CU to a CTU |
| 103 | class CUData |
| 104 | { |
| 105 | public: |
| 106 | |
| 107 | static cubcast_t s_partSet[NUM_FULL_DEPTH]; // pointer to broadcast set functions per absolute depth |
| 108 | static uint32_t s_numPartInCUSize; |
| 109 | |
| 110 | FrameData* m_encData; |
| 111 | const Slice* m_slice; |
| 112 | |
| 113 | cucopy_t m_partCopy; // pointer to function that copies m_numPartitions elements |
| 114 | cubcast_t m_partSet; // pointer to function that sets m_numPartitions elements |
| 115 | cucopy_t m_subPartCopy; // pointer to function that copies m_numPartitions/4 elements, may be NULL |
| 116 | cubcast_t m_subPartSet; // pointer to function that sets m_numPartitions/4 elements, may be NULL |
| 117 | |
| 118 | uint32_t m_cuAddr; // address of CTU within the picture in raster order |
| 119 | uint32_t m_absIdxInCTU; // address of CU within its CTU in Z scan order |
| 120 | uint32_t m_cuPelX; // CU position within the picture, in pixels (X) |
| 121 | uint32_t m_cuPelY; // CU position within the picture, in pixels (Y) |
| 122 | uint32_t m_numPartitions; // maximum number of 4x4 partitions within this CU |
| 123 | |
| 124 | int m_chromaFormat; |
| 125 | int m_hChromaShift; |
| 126 | int m_vChromaShift; |
| 127 | |
| 128 | /* Per-part data, stored contiguously */ |
| 129 | char* m_qp; // array of QP values |
| 130 | uint8_t* m_log2CUSize; // array of cu log2Size TODO: seems redundant to depth |
| 131 | uint8_t* m_partSize; // array of partition sizes |
| 132 | uint8_t* m_predMode; // array of prediction modes |
| 133 | uint8_t* m_lumaIntraDir; // array of intra directions (luma) |
| 134 | uint8_t* m_tqBypass; // array of CU lossless flags |
| 135 | char* m_refIdx[2]; // array of motion reference indices per list |
| 136 | uint8_t* m_cuDepth; // array of depths |
| 137 | uint8_t* m_skipFlag; // array of skip flags |
| 138 | uint8_t* m_mergeFlag; // array of merge flags |
| 139 | uint8_t* m_interDir; // array of inter directions |
| 140 | uint8_t* m_mvpIdx[2]; // array of motion vector predictor candidates or merge candidate indices [0] |
| 141 | uint8_t* m_tuDepth; // array of transform indices |
| 142 | uint8_t* m_transformSkip[3]; // array of transform skipping flags per plane |
| 143 | uint8_t* m_cbf[3]; // array of coded block flags (CBF) per plane |
| 144 | uint8_t* m_chromaIntraDir; // array of intra directions (chroma) |
| 145 | enum { BytesPerPartition = 22 }; // combined sizeof() of all per-part data |
| 146 | |
| 147 | coeff_t* m_trCoeff[3]; // transformed coefficient buffer per plane |
| 148 | |
| 149 | MV* m_mv[2]; // array of motion vectors per list |
| 150 | MV* m_mvd[2]; // array of coded motion vector deltas per list |
| 151 | enum { TMVP_UNIT_MASK = 0xF0 }; // mask for mapping index to into a compressed (reference) MV field |
| 152 | |
| 153 | const CUData* m_cuAboveLeft; // pointer to above-left neighbor CTU |
| 154 | const CUData* m_cuAboveRight; // pointer to above-right neighbor CTU |
| 155 | const CUData* m_cuAbove; // pointer to above neighbor CTU |
| 156 | const CUData* m_cuLeft; // pointer to left neighbor CTU |
| 157 | |
| 158 | CUData(); |
| 159 | |
| 160 | void initialize(const CUDataMemPool& dataPool, uint32_t depth, int csp, int instance); |
| 161 | void calcCTUGeoms(uint32_t picWidth, uint32_t picHeight, uint32_t maxCUSize, CUGeom cuDataArray[CUGeom::MAX_GEOMS]) const; |
| 162 | |
| 163 | void initCTU(const Frame& frame, uint32_t cuAddr, int qp); |
| 164 | void initSubCU(const CUData& ctu, const CUGeom& cuGeom); |
| 165 | void initLosslessCU(const CUData& cu, const CUGeom& cuGeom); |
| 166 | |
| 167 | void copyPartFrom(const CUData& cu, const CUGeom& childGeom, uint32_t subPartIdx); |
| 168 | void setEmptyPart(const CUGeom& childGeom, uint32_t subPartIdx); |
| 169 | void copyToPic(uint32_t depth) const; |
| 170 | |
| 171 | /* RD-0 methods called only from encodeResidue */ |
| 172 | void copyFromPic(const CUData& ctu, const CUGeom& cuGeom); |
| 173 | void updatePic(uint32_t depth) const; |
| 174 | |
| 175 | void setPartSizeSubParts(PartSize size) { m_partSet(m_partSize, (uint8_t)size); } |
| 176 | void setSkipFlagSubParts(uint8_t skipFlag) { m_partSet(m_skipFlag, skipFlag); } |
| 177 | void setPredModeSubParts(PredMode mode) { m_partSet(m_predMode, (uint8_t)mode); } |
| 178 | void clearCbf() { m_partSet(m_cbf[0], 0); m_partSet(m_cbf[1], 0); m_partSet(m_cbf[2], 0); } |
| 179 | |
| 180 | /* these functions all take depth as an absolute depth from CTU, it is used to calculate the number of parts to copy */ |
| 181 | void setQPSubParts(char qp, uint32_t absPartIdx, uint32_t depth) { s_partSet[depth]((uint8_t*)m_qp + absPartIdx, (uint8_t)qp); } |
| 182 | void setTUDepthSubParts(uint8_t tuDepth, uint32_t absPartIdx, uint32_t depth) { s_partSet[depth](m_tuDepth + absPartIdx, tuDepth); } |
| 183 | void setLumaIntraDirSubParts(uint8_t dir, uint32_t absPartIdx, uint32_t depth) { s_partSet[depth](m_lumaIntraDir + absPartIdx, dir); } |
| 184 | void setChromIntraDirSubParts(uint8_t dir, uint32_t absPartIdx, uint32_t depth) { s_partSet[depth](m_chromaIntraDir + absPartIdx, dir); } |
| 185 | void setCbfSubParts(uint8_t cbf, TextType ttype, uint32_t absPartIdx, uint32_t depth) { s_partSet[depth](m_cbf[ttype] + absPartIdx, cbf); } |
| 186 | void setCbfPartRange(uint8_t cbf, TextType ttype, uint32_t absPartIdx, uint32_t coveredPartIdxes) { memset(m_cbf[ttype] + absPartIdx, cbf, coveredPartIdxes); } |
| 187 | void setTransformSkipSubParts(uint8_t tskip, TextType ttype, uint32_t absPartIdx, uint32_t depth) { s_partSet[depth](m_transformSkip[ttype] + absPartIdx, tskip); } |
| 188 | void setTransformSkipPartRange(uint8_t tskip, TextType ttype, uint32_t absPartIdx, uint32_t coveredPartIdxes) { memset(m_transformSkip[ttype] + absPartIdx, tskip, coveredPartIdxes); } |
| 189 | |
| 190 | bool setQPSubCUs(char qp, uint32_t absPartIdx, uint32_t depth); |
| 191 | |
| 192 | void setPUInterDir(uint8_t dir, uint32_t absPartIdx, uint32_t puIdx); |
| 193 | void setPUMv(int list, const MV& mv, int absPartIdx, int puIdx); |
| 194 | void setPURefIdx(int list, char refIdx, int absPartIdx, int puIdx); |
| 195 | |
| 196 | uint8_t getCbf(uint32_t absPartIdx, TextType ttype, uint32_t trDepth) const { return (m_cbf[ttype][absPartIdx] >> trDepth) & 0x1; } |
| 197 | uint8_t getQtRootCbf(uint32_t absPartIdx) const { return m_cbf[0][absPartIdx] || m_cbf[1][absPartIdx] || m_cbf[2][absPartIdx]; } |
| 198 | char getRefQP(uint32_t currAbsIdxInCTU) const; |
| 199 | uint32_t getInterMergeCandidates(uint32_t absPartIdx, uint32_t puIdx, MVField (*mvFieldNeighbours)[2], uint8_t* interDirNeighbours) const; |
| 200 | void clipMv(MV& outMV) const; |
| 201 | int fillMvpCand(uint32_t puIdx, uint32_t absPartIdx, int picList, int refIdx, MV* amvpCand, MV* mvc) const; |
| 202 | void getIntraTUQtDepthRange(uint32_t tuDepthRange[2], uint32_t absPartIdx) const; |
| 203 | void getInterTUQtDepthRange(uint32_t tuDepthRange[2], uint32_t absPartIdx) const; |
| 204 | |
| 205 | uint32_t getNumPartInter() const { return nbPartsTable[(int)m_partSize[0]]; } |
| 206 | bool isIntra(uint32_t absPartIdx) const { return m_predMode[absPartIdx] == MODE_INTRA; } |
| 207 | bool isSkipped(uint32_t absPartIdx) const { return !!m_skipFlag[absPartIdx]; } |
| 208 | bool isBipredRestriction() const { return m_log2CUSize[0] == 3 && m_partSize[0] != SIZE_2Nx2N; } |
| 209 | |
| 210 | void getPartIndexAndSize(uint32_t puIdx, uint32_t& absPartIdx, int& puWidth, int& puHeight) const; |
| 211 | void getMvField(const CUData* cu, uint32_t absPartIdx, int picList, MVField& mvField) const; |
| 212 | |
| 213 | void getAllowedChromaDir(uint32_t absPartIdx, uint32_t* modeList) const; |
| 214 | int getIntraDirLumaPredictor(uint32_t absPartIdx, uint32_t* intraDirPred) const; |
| 215 | void deriveLeftRightTopIdxAdi(uint32_t& partIdxLT, uint32_t& partIdxRT, uint32_t partOffset, uint32_t partDepth) const; |
| 216 | |
| 217 | uint32_t getSCUAddr() const { return (m_cuAddr << g_maxFullDepth * 2) + m_absIdxInCTU; } |
| 218 | uint32_t getCtxSplitFlag(uint32_t absPartIdx, uint32_t depth) const; |
| 219 | uint32_t getCtxSkipFlag(uint32_t absPartIdx) const; |
| 220 | ScanType getCoefScanIdx(uint32_t absPartIdx, uint32_t log2TrSize, bool bIsLuma, bool bIsIntra) const; |
| 221 | void getTUEntropyCodingParameters(TUEntropyCodingParameters &result, uint32_t absPartIdx, uint32_t log2TrSize, bool bIsLuma) const; |
| 222 | |
| 223 | const CUData* getPULeft(uint32_t& lPartUnitIdx, uint32_t curPartUnitIdx) const; |
| 224 | const CUData* getPUAbove(uint32_t& aPartUnitIdx, uint32_t curPartUnitIdx, bool planarAtCTUBoundary = false) const; |
| 225 | const CUData* getPUAboveLeft(uint32_t& alPartUnitIdx, uint32_t curPartUnitIdx) const; |
| 226 | const CUData* getPUAboveRight(uint32_t& arPartUnitIdx, uint32_t curPartUnitIdx) const; |
| 227 | const CUData* getPUBelowLeft(uint32_t& blPartUnitIdx, uint32_t curPartUnitIdx) const; |
| 228 | |
| 229 | const CUData* getQpMinCuLeft(uint32_t& lPartUnitIdx, uint32_t currAbsIdxInCTU) const; |
| 230 | const CUData* getQpMinCuAbove(uint32_t& aPartUnitIdx, uint32_t currAbsIdxInCTU) const; |
| 231 | |
| 232 | const CUData* getPUAboveRightAdi(uint32_t& arPartUnitIdx, uint32_t curPartUnitIdx, uint32_t partUnitOffset) const; |
| 233 | const CUData* getPUBelowLeftAdi(uint32_t& blPartUnitIdx, uint32_t curPartUnitIdx, uint32_t partUnitOffset) const; |
| 234 | |
| 235 | protected: |
| 236 | |
| 237 | template<typename T> |
| 238 | void setAllPU(T *p, const T& val, int absPartIdx, int puIdx); |
| 239 | |
| 240 | char getLastCodedQP(uint32_t absPartIdx) const; |
| 241 | int getLastValidPartIdx(int absPartIdx) const; |
| 242 | |
| 243 | bool hasEqualMotion(uint32_t absPartIdx, const CUData& candCU, uint32_t candAbsPartIdx) const; |
| 244 | |
| 245 | bool isDiffMER(int xN, int yN, int xP, int yP) const; |
| 246 | |
| 247 | // add possible motion vector predictor candidates |
| 248 | bool addMVPCand(MV& mvp, int picList, int refIdx, uint32_t absPartIdx, MVP_DIR dir) const; |
| 249 | bool addMVPCandOrder(MV& mvp, int picList, int refIdx, uint32_t absPartIdx, MVP_DIR dir) const; |
| 250 | |
| 251 | bool getColMVP(MV& outMV, int& outRefIdx, int picList, int cuAddr, int absPartIdx) const; |
| 252 | |
| 253 | void scaleMvByPOCDist(MV& outMV, const MV& inMV, int curPOC, int curRefPOC, int colPOC, int colRefPOC) const; |
| 254 | |
| 255 | void deriveLeftRightTopIdx(uint32_t puIdx, uint32_t& partIdxLT, uint32_t& partIdxRT) const; |
| 256 | |
| 257 | uint32_t deriveCenterIdx(uint32_t puIdx) const; |
| 258 | uint32_t deriveRightBottomIdx(uint32_t puIdx) const; |
| 259 | uint32_t deriveLeftBottomIdx(uint32_t puIdx) const; |
| 260 | }; |
| 261 | |
| 262 | // TU settings for entropy encoding |
| 263 | struct TUEntropyCodingParameters |
| 264 | { |
| 265 | const uint16_t *scan; |
| 266 | const uint16_t *scanCG; |
| 267 | ScanType scanType; |
| 268 | uint32_t log2TrSizeCG; |
| 269 | uint32_t firstSignificanceMapContext; |
| 270 | }; |
| 271 | |
| 272 | struct CUDataMemPool |
| 273 | { |
| 274 | uint8_t* charMemBlock; |
| 275 | coeff_t* trCoeffMemBlock; |
| 276 | MV* mvMemBlock; |
| 277 | |
| 278 | CUDataMemPool() { charMemBlock = NULL; trCoeffMemBlock = NULL; mvMemBlock = NULL; } |
| 279 | |
| 280 | bool create(uint32_t depth, uint32_t csp, uint32_t numInstances) |
| 281 | { |
| 282 | uint32_t numPartition = NUM_CU_PARTITIONS >> (depth * 2); |
| 283 | uint32_t cuSize = g_maxCUSize >> depth; |
| 284 | uint32_t sizeL = cuSize * cuSize; |
| 285 | uint32_t sizeC = sizeL >> (CHROMA_H_SHIFT(csp) + CHROMA_V_SHIFT(csp)); |
| 286 | CHECKED_MALLOC(trCoeffMemBlock, coeff_t, (sizeL + sizeC * 2) * numInstances); |
| 287 | CHECKED_MALLOC(charMemBlock, uint8_t, numPartition * numInstances * CUData::BytesPerPartition); |
| 288 | CHECKED_MALLOC(mvMemBlock, MV, numPartition * 4 * numInstances); |
| 289 | return true; |
| 290 | |
| 291 | fail: |
| 292 | return false; |
| 293 | } |
| 294 | |
| 295 | void destroy() |
| 296 | { |
| 297 | X265_FREE(trCoeffMemBlock); |
| 298 | X265_FREE(mvMemBlock); |
| 299 | X265_FREE(charMemBlock); |
| 300 | } |
| 301 | }; |
| 302 | } |
| 303 | |
| 304 | #endif // ifndef X265_CUDATA_H |